AI Hacking Defense: White House vs. OpenAI Debate

A new AI called Mythos has found thousands of high-severity vulnerabilities in major operating systems, web browsers, and core cryptography libraries. This discovery has sparked a debate about how to best defend against AI hacking. The White House opposes expanding access to Mythos due to security risks. It's even considering requiring AI companies to vet frontier models with the government before release. This marks a potential shift from its previous light-touch regulatory approach. Companies like OpenAI argue that attackers won't wait. They say criminal groups will adopt these tools regardless, and the best defense is to equip trusted defenders faster than adversaries can adapt. OpenAI calls this "controlled acceleration," with a tiered access program for vetted defenders. This debate challenges traditional cybersecurity standards, where vulnerabilities are responsibly disclosed after a patch is ready. AI models like Mythos can find vulnerabilities so quickly that the window for responsible coordination narrows significantly. Currently, Anthropic, the creator of Mythos, has restricted access to a limited private release, coordinating quietly with a small group. This means the company is effectively setting the terms of disclosure and access. This situation matters because government agencies are rapidly adopting AI tools, creating a gap between their use and regulatory oversight.